Findings based on a large retrospective review of screening photographs
THURSDAY, Sept. 21, 2017 (HealthDay News) — Deep neural network or artificial intelligence is able to detect referable diabetic retinopathy from photographs, according to research published online Sept. 4 in Clinical & Experimental Ophthalmology.
Nishanthan Ramachandran, M.B.Ch.B., from Dunedin Hospital in New Zealand, and colleagues retrospectively reviewed diabetic retinal photos from the Otago database, photographed during October 2016 (485 photos), as well as 1,200 photos from the Messidor international database. The team used a third party deep neural network software to assess diabetic retinopathy grades of the photos.
The researchers found that for detecting referable diabetic retinopathy, the deep neural network had an area under the receiver operating characteristic curve of 0.901, with 84.6 percent sensitivity and 79.7 percent specificity for the Otago photos. For the Messidor photos, the area under the receiver operating characteristic curve was 0.980, with 96.0 percent sensitivity and 90.0 percent specificity.
“We believe that deep neural networks can be integrated into community screening once they can successfully detect both diabetic retinopathy and diabetic macular edema,” the authors write.
Abstract
Full Text (subscription or payment may be required)
Copyright © 2017 HealthDay. All rights reserved.
